How to calculate rolling bootstrapped values and confidence intervals in R

Question

I am new to R and am trying to calculate the bootstrapped standard deviation (sd) and associated standard error within a 30 observation rolling window. The function below performs the rolling window appropriately if I just want sd. But when I add the bootstrap function using the boot package I get the error specified below. I gather that I am trying to store bootstrap results in a vector that isn't the correct size. Does anyone have any advice on how to store just the bootstrapped sd and associated stderror for each window in rows of a new matrix? The goal is to then plot the sd and associated 95% confidence intervals for each window along the timeseries. Thanks in advance for any help.

> head(data.srs)
    LOGFISH
1  0.8274083
2  1.0853433
3  0.8049845
4  0.8912097
5  1.3514569
6  0.8694499


###Function to apply rolling window

rollWin <- function(timeSeries,  windowLength) 
{
  data<-timeSeries
  nOut <- length(data[, 1]) - windowLength + 1
  out <- numeric(nOut)
  if (length(data[,1]) >= windowLength)
  {
    for (i in 1:nOut) 
      { 
      sd.fun <- function(data,d)sd(data[d], na.rm = TRUE)
      out[i] <- boot(data[i:(i + windowLength - 1), ], sd.fun, R=1000)
      }
  }
  return (list(result=out))
} 

###run rolling window function. ex. rollWin(data, windowlength)
a.temp<-rollWin(data.srs,30)


> warnings()
Warning messages:
1: In out[i] <- boot(data[i:(i + windowLength - 1), ], sd.fun,  ... :
  number of items to replace is not a multiple of replacement length

Answer 1

You can simplify it quite a lot. I am not familiar with the boot package, but we can roll a function along a vector using the rollapply function quite easily, and then we can make bootstrap samples using the replicate function:

# Create some data, 12 items long
r <- runif(12)
# [1] 0.44997964 0.27425412 0.07327872 0.68054759 0.33577348 0.49239478
# [7] 0.93421646 0.19633079 0.45144966 0.53673296 0.71813017 0.85270346


require(zoo)

# use rollapply to calculate function alonga  moving window
# width is the width of the window
sds <- rollapply( r , width = 4 , by = 1 , sd )
#[1] 0.19736258 0.26592331 0.16770025 0.12585750 0.13730946 0.08488467
#[7] 0.16073722 0.22460430 0.22462168


# Now we use replicate to repeatedly evaluate a bootstrap sampling method
# 'n' is number of replications
n <- 4
replicate( n , rollapply( r , width = n , function(x) sd( x[ sample(length(x) , repl = TRUE) ] ) ) )



#            [,1]      [,2]       [,3]      [,4]
# [1,] 0.17934073 0.1815371 0.11603320 0.2992379
# [2,] 0.03551822 0.2862702 0.18492837 0.2526193
# [3,] 0.09042535 0.2419768 0.13124738 0.1666012
# [4,] 0.17238705 0.1410475 0.18136178 0.2457248
# [5,] 0.32008385 0.1709326 0.32909368 0.2550859
# [6,] 0.30832533 0.1480320 0.02363968 0.1275594
# [7,] 0.23069951 0.1275594 0.25648052 0.3016909
# [8,] 0.11235170 0.2493055 0.26089969 0.3012610
# [9,] 0.16819174 0.2099518 0.18033502 0.0906986

Each column represents the rollapply which bootstraps the observations in the current window before applying sd.